Toots for dpiponi@mathstodon.xyz account

Written by Dan Piponi on 2024-11-24 at 01:28

If you replace die() with a standard normal variate you tend to get around -0.3. I wonder how hard it is to compute the exact result for the limit - assuming there is a limit.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-24 at 01:02

Code https://godbolt.org/z/Thex6n8ev

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-24 at 00:44

I was wondering how much you could control the resulting distribution if the only tools you have available are 6-sided dice and the functions min and max. Looks like you can keep "refining" the distribution to get it as close as you like to being supported at just one value.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-24 at 00:42

I think this is curious. die() rolls a die. This code almost always returns 3 for large input n.

int depth(int n)

{

if (n == 0)

{

return die(6);

}

return max(

min(depth(n - 1), depth(n - 1)),

min(depth(n - 1), depth(n - 1)));

}

(I have Haskell code to compute the exact prob.)

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-17 at 21:59

Wondering about the masking used in the attention mechanism.

Don't you end up putting way more work into training for the case when the context isn't "full" even though in reality you only need that for some initial prefix, after which you're always generating with a full context?

Or am I misunderstanding something?

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-16 at 16:34

Musk bought Twitter as a platform for his own voice. He succeeded and has now worked his way into the cesspit called the government of the US. At which point what does he care if you leave Twitter. He's already succeeded. I find the exodus rather pathetic.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-15 at 21:38

I think Gandalf was mean when he insisted on Frodo and Sam wearing old rags, and then made Frodo wear a sword. I can understand why, at this point, everyone else might fetishize those things. But Frodo and Sam's wishes should have come first.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-15 at 15:36

This is the American Dream where no level of ignorance and stupidity is a barrier to success.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-15 at 02:48

I got to the bit where they mentioned en- and em-dashes and realised that this is where I get my classic ebooks from in future. Even the for-money publishers don’t seem to understand dashes these days.

https://standardebooks.org

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-15 at 00:42

Voice recognition fully integrated into the operating system is good ‘n’ all. Until you realize that you have a whole load of directories whose names end in a space.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-14 at 15:59

I rarely interact with cmake but I thought I'd try again to learn it even though I'm not entirely sure what purpose it serves. So cmake 101 here...

So I have my fresh new raspberry-pi with everything installed in its default place. I have a small ncurses app. I follow the instructions on how to add curses to a project. It can't find the libraries. I look for help online. I'm told to look for the libraries myself and add the locations to the CMakeLists.txt file. I find the libraries in a raspberry-pi specific location. Now it works. So what did cmake buy me here if I still need to write O($NUM_PLATFORMS*$NUM_LIBRARIES) lines of code?

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-14 at 15:06

Clown country is clowning pretty hard with its clown cabinet. I'd better get my UK passport up to date.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-14 at 05:29

Hard to reconcile that with being told there were rapidly dividing cells but apparently it's the quality of the nuclei that's more important.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-14 at 05:14

Cone cat's tumour is apparently benign.

=> View attached media

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-14 at 03:22

Oh well, I guess it had to happen eventually @dpiponi.bsky.social

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-13 at 17:03

The instrument panel in the car this morning. Pulled over to the side of the road to reboot.

=> View attached media

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-12 at 20:39

The reasoning that leads to using Dropout in machine learning suggests that you should probably only ever study while drunk.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-12 at 17:35

I found a video here:

https://www.youtube.com/watch?v=pKq4nD9TO9w

In my runs, blinkers dominate due to changing state every single frame. As I don't see any effects from blinkers in that video it can't be the same thing. It's similar though.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-12 at 15:43

BTW I wrote this code using colab's AI. I literally typed in something like "implement the game of life showing how many times each grid cell changes value with a heat map" and had to fix one minor bug relating to the colour map.

=> More informations about this toot | View the thread

Written by Dan Piponi on 2024-11-12 at 02:05

The original image has faint criss-crossing glider paths but it may have been lost when stored compressed on Mathstodon.

=> More informations about this toot | View the thread

=> This profile with reblog | Go to dpiponi@mathstodon.xyz account

Proxy Information
Original URL
gemini://mastogem.picasoft.net/profile/109303820089311192
Status Code
Success (20)
Meta
text/gemini
Capsule Response Time
333.561401 milliseconds
Gemini-to-HTML Time
4.47033 milliseconds

This content has been proxied by September (ba2dc).